Relational Machine Learning for Knowledge Graphs


Ivana Balazevic

Knowledge bases (KBs), such as Google Knowledge Graph, Wordnet [1], and Freebase [2] are important information resources, containing facts about the real world in the formof a large knowledge graph.  Each fact in the knowledge base is represented as a resource description framework (RDF) triplet: (subject, predicate, object), where subject and object are considered entities represented as nodes in the graph, and predicate is considered as a relation between those entities represented as a directed edge between the nodes.

Knowledge graphs are used for a wide variety of natural language processing tasks, e.g. question answering, information retrieval, co-reference resolution, etc.  However, the problem with such resources is that they are incomplete and lack reasoning capability, which inspired our work in this area.


Supervisors: Tim Hospedales & Ivan Titov


Previous: Sampling Methods for Speech Synthesis: Using Generative Models to vary Prosody